Persisted before performed
Plans are written down before they run, so work survives a restart and can be reasoned about.
Development
Built in the open, honestly.
Most products show you the demo. Morrow shows you the build. Here is the real roadmap, the current state, and how we intend to prove what we claim.
Current milestone In progress
First reliable vertical slice
One complete path from request to verified result.
- Persisted task lifecycle with clear states
- Task events, checkpoints, and restart recovery
- A connected interface over the real engine
The roadmap
Eight stages, in order.
Each stage depends on the one before it. We would rather land one stage reliably than fake five.
- Stage 1 Complete
Foundation
The ground the rest of Morrow stands on.
- Repository, versioned contracts, and local tooling
- SQLite migration foundation
- Basic project and task persistence
- Stage 2 In progress
First reliable vertical slice
One complete path from request to verified result.
- Persisted task lifecycle with clear states
- Task events, checkpoints, and restart recovery
- A connected interface over the real engine
- Stage 3 Planned
Secure tool execution
Actions that are scoped, visible, and reversible.
- Workspace-scoped file inspection
- Permission boundaries per action
- Evidence retained for every meaningful step
- Stage 4 Planned
Model providers
Your choice of local or approved remote models.
- Local inference targets
- Approved provider adapters
- Per-task model routing
- Stage 5 Planned
Structured memory
Memory that is deliberate, scoped, and inspectable.
- Typed memory with scopes
- Retention controls
- Visible reads and writes
- Stage 6 Planned
Persistent agent teams
Named agents with roles, limits, and history.
- Roles, models, and tools per agent
- Cost and permission limits
- Activity history
- Stage 7 Planned
Automations and integrations
Triggers and connections, under explicit boundaries.
- Event hooks and schedules
- Connected services
- Scoped, auditable access
- Stage 8 Research
Compatibility and benchmarks
Claims backed by repeatable, public tests.
- Compatibility work
- Reproducible benchmark suite
- Published methodology and results
Status, today
What exists, what’s moving, what’s next.
Complete 6
- Repository foundation
- Product and architecture documentation
- Local development tooling
- Versioned contracts
- SQLite migration foundation
- Basic project and task persistence
In progress 4
- Persisted task lifecycle
- Secure workspace inspection
- Task events and recovery
- Connected product interface
Planned 9
- Model providers
- Browser and terminal tools
- Structured memory
- Persistent agent teams
- Automations
- Integrations
- Desktop application
- Compatibility work
- Public benchmark suite
Architecture principles
The rules the build holds itself to.
Evidence over assertion
Every meaningful action is meant to leave an artifact you can inspect after the fact.
Contracts, then code
Versioned contracts define behavior first; implementations are held to them.
Reversible by default
Where an action can be undone, it should be — and interrupted work is shown honestly.
Benchmark philosophy
Claims will be backed by repeatable tests — not curated demos.
Morrow is being measured against a public benchmark suite that anyone can run. Until a capability passes a reproducible test, we will describe it as a goal, not a result. When it passes, we intend to publish the methodology alongside the numbers.
Source
Following the build.
The public source repository will be linked here when it opens. Until then, join early access to get the announcement directly.
Join early access →Watch it take shape.
This page is kept current as the build moves. Bookmark it, or join early access to follow along.